Effective ITS and traffic management purposes requires a complete and accurate information about current and predicted traffic states in the transport network. The current state-of-the-art in literature regarding traffic state estimation and prediction yields efforts which mostly focus on highways, which are not bluntly transferrable to an urban environment and do not maximize the utilization of all available traffic data. This paper describes the development and assessment of a data-driven traffic state estimation and prediction framework for application in an urban environment. It uses the intuitive relationship between past, current and future traffic states on neighboring links to train and improve estimation/prediction accuracy and fill the gaps on those links where no floating car data are available. Additionally, this framework is tested on the well-known Sioux Falls Scenario. When penetration rate of floating cars is 5%, on average 50% of the urban links are estimated within 5 km/h accuracy. For a prediction horizon of 5 minutes, it performs almost equal with a percentage of 49%.